A neural network-based machine vision method for surface roughness measurement

In our current study, a neural network-based machine vision method is proposed to measure the surfaces roughness for different ¿38 mm grinding shafts in different ambient light conditions. Firstly, the effect of ambient light is analyzed using the two approaches, i.e., the approach of standard deviation of gray-level distribution proposed by Luk and that based on gray-level co-occurrence matrix. Then, a new RBF neural network-based method is proposed to measure the roughness by extracting the features of ambient light and work piece. The neural network is trained by five work pieces with known surface roughness, and eleven work pieces are tested by the proposed method. An analytical comparison between the proposed method and the two existing ones mentioned above verifies that our method is of better performance with least variance sum.